The Polarization of Information on the Web Charles Dickens - PowerPoint PPT Presentation

The Polarization of Information on the Web Charles Dickens Crystal Harper Cook Ram Hari Dahal Prajjwal Dangal Pamela Bilo Thomas

Objective & Impact • The primary goal of this study is to compare the polarization of opinions, information, and users on the web • Given the current polarized political climate in the United States, a better understanding of how people view certain issues and how those viewpoints may be connected to other topics could help bridge the gap and facilitate discourse between groups.

Approach Our focus for the week was primarily on Twitter. 1) We will develop a fully connected directed graph where Tweets are represented as nodes of a graph and edge weights correspond to the transition probabilities between the nodes.

Approach Our focus for the week was primarily on Twitter. 1) We will develop a fully connected directed graph where Tweets are represented as nodes of a graph and edge weights correspond to the transition probabilities between the nodes. 2) The graph will then be input to our clustering algorithm to identify communities of thought.

Approach Our focus for the week was primarily on Twitter. 1) We will develop a fully connected directed graph where Tweets are represented as nodes of a graph and edge weights correspond to the transition probabilities between the nodes. 2) The graph will then be input to our clustering algorithm to identify communities of thought. 3) The polarization of our sampled Tweets can then be compared based on the average conductance of the retrieved clusters.

Data Collection Source: https://data.world/crowdflower/progressive-issues-sentiment Four topics: Abortion, Atheism, Hillary Clinton, Feminism Three outcomes: For, Against, Neutral Source: https://data.world/bkey/politician-tweets Collect hashtags tweeted by politicians and compare how many hashtags are used by both

Weight Calculation - Sentiment For each tweet we are given the sentiment classification: - For or Against Along with a confidence value with the range 0 and 1 Against tweets had their confidence values multiplied by -1 Edge weights are calculated based on their distance

Weight Calculation - Political Hashtags For each pair of politicians: Calculate how many hashtags the two politicians have in common Normalize on a scale of 0-1 based upon the most hashtags shared between two politicians

Results Topic: Atheism Our Model’s Conductance Calculations Sentiment Only Conductance Score: 0.5029761904761905

Results Topic: Feminist Movement Sentiment Only Conductance Score: 0.5021097046413502

Results Topic: Hillary Clinton Sentiment Only Conductance Score: 0.502283105022831

Results Topic: Legalization of Abortion Sentiment Only Conductance Score: 0.5024630541871922

Results Topic: Politician All Hashtags

Results Topic: Politician Filtered Hashtags

Visualization

Further work A publicly available web app: ● Allows users to choose from a range of current issues ● Include substantial amount of data for each topic ● Include other data sources (Facebook posts, comments on news articles)

Further work cont. ● Polarization trend over time for the same topic ● Incorporate features such as a tweet’s popularity, a user’s popularity, and user interactions into edge weight calculation ● Inter-issue similarity ● Add more types of visualization

References Data: Progressive Issue Sentiment Analysis https://www.figure-eight.com/data-for-everyone/ Data: Politician Tweets https://data.world/bkey/politician-tweets

Thank You Kelly Andronicos Brent Ladd Mark Ward Tsai-Wei Wu Carolyn Johnson Tyler Netherly Elizabeth Bell Yucong Zhang